Prompt research
Search an index of Google AI Overview and ChatGPT answers to find the prompts your buyers ask, who the answers name, and where the gap is
Daily tracking measures the prompts you already track. Research looks outward: which prompts about your category exist, which of them name a competitor and not you, which sites the answers lean on, and how often each prompt is asked. It reads an index of Google AI Overview and ChatGPT answers. It never fetches a live answer.
Open Research in the property navigation. Every member with access to the property can run a Search, on every plan, within the allowance below.
The page
The cover carries the title, one line on what the page reads, and the remaining allowance. Under it sits a toolbar that stays put whatever is open: the search input, the Market picker, the index toggle (Google AI Overview or ChatGPT), three kind pills labelled Prompts, Mentions and Citations, and the Search button.
Each result opens as a tab beside the toolbar, labelled with its target and kind. Several tabs can stay open at once, and a Search still reading the index keeps going while you read another tab. The History button opens a sheet listing every Search the property has run; opening a row adds a tab.
What the index is
The index is a record of prompts that Google AI Overview and ChatGPT have been observed answering. For each prompt it holds the answer, the sites the answer cited, the brands it named, the follow-up prompts it fanned out to, and an estimate of how often the prompt is asked of AI.
A Search looks rows up in that index. It does not send anything to ChatGPT or Google, and the answer you read is the one the index recorded, not a fresh one. The page says so in one line above every result. Heading's own measurement of your prompts is daily tracking, which is a different thing: see How prompts work.
Three kinds of Search
A Search is one kind, one target, one Market and one index.
| Kind | What you type | What comes back | Example |
|---|---|---|---|
| Topic | A phrase | Prompts whose question contains the phrase | running shoes for flat feet |
| Mention | A brand name | Prompts whose answer names the brand | Nike |
| Citation | A site | Prompts whose answer cites the site among its sources | nike.com |
The pills in the toolbar name the three kinds by what comes back: Prompts runs a Topic Search, Mentions a Mention Search and Citations a Citation Search. Type something that looks like a site while Prompts is selected and the page offers to run it under Citations instead.
Before the first result is open, the page shows quick-start chips for your own position: a Mention Search for your brand, a Citation Search for your site, and a Mention Search for each competitor you track. Each chip states whether it is cached or will use one search, so the cost of a click is never a surprise. Once a tab is open the chips are gone and the toolbar is the way in.
What one Search returns
The 200 most-asked indexed prompts that match, ordered by AI search volume. Every figure and every roll-up on the page covers those 200 and its caption says so. When the index holds more matches than that, the first figure states how many.
A result tab has three views, named as the pills are: Indexed prompts (the table), Mentions (who the answers name) and Citations (the sites the answers cite). The provenance line, the fetch day and the observation period sit behind the info dot beside the views.
The four figures
| Figure | What it counts |
|---|---|
| Indexed prompts | The 200 most-asked prompts that match, and how many the index holds in total |
| AI search volume | The index's estimate of how often those 200 prompts are asked of AI in a month, summed and rounded |
| Name {your brand} | How many of the 200 answers name your brand, matched the same way daily tracking matches a mention |
| Cite {your site} | How many of the 200 answers list your site among their sources |
The two leaderboards
- Who the answers name: your brand and each tracked competitor, with how many of the 200 answers name it, how many cite it, and its share of the volume. On the Google index only brands you track can appear here. On the ChatGPT index a second list shows the other brands the index named, which is one way to find a competitor you do not track yet.
- Sites the answers cite: the 10 sites cited most often across the 200 answers, with your own site and your competitors' sites marked.
The Prompt gap filter
The Indexed prompts view has a filter dropdown with three options: All, {your brand} appears, and Prompt gap. The Prompt gap is every prompt where a competitor is named or cited and your brand is not. It is the list to read first.
The row detail
Each row shows the prompt, its AI search volume on the same 1 to 5 bar the prompts table uses, the tracked brands its answer names, and how many sources the answer cites. Open a row to read the whole answer, the sources with links, the prompts the answer fanned out to, and the volume by month.
A prompt the index holds no volume estimate for reads "No estimate", never 0. Absent is not zero.
Freshness and refresh
A result is served from cache for 30 days. Inside that window, running the same Search again opens the cached result and costs nothing, whoever ran it first and whichever team they were on.
After 7 days a Refresh control appears on the result. Under 7 days it is absent rather than disabled, because a refresh inside a week buys nothing. A refresh reads the index again, counts as one search, and creates a new version of the result. Earlier versions are kept.
Every result states the day it was fetched and the period in which its answers were observed.
Comparing two versions
When a Search has a previous version, Compare with the previous fetch lists three things:
- prompts that are new to the top 200
- prompts that left the top 200
- prompts whose answer now names a different set of brands
The comparison describes a change in the index. The index is rebuilt on a schedule that is not published, so a prompt leaving the top 200, or a brand leaving an answer, is the index changing. It is not your visibility moving, and the page never says it is. For the same reason there is no trend chart over versions. Your brand's own movement is what daily tracking measures.
The allowance
A Search that reads the index is a fetching search, and each property has an allowance of them per rolling 30 days.
| Plan | Fetching searches per property, per rolling 30 days |
|---|---|
| Free (and lapsed) | 3 |
| Lite | 6 |
| Starter | 12 |
| Growth | 30 |
| Scale | 60 |
- A cached result never counts.
- A refresh counts as one search.
- A fetch that fails is not counted.
- The window rolls: a search comes back 30 days after it was used.
The remaining allowance is shown in the cover. Once it is used up, cached results still open, and a quick-start chip that is cached still works. The Search button states the allowance instead of going quiet.
Adding an indexed prompt to your portfolio
Select Add to portfolio on one row, or tick several rows and select Add N to portfolio.
- The prompt is created active, in the same Market as the Search.
- It counts against the property's prompt quota like any other prompt. See Plans.
- It carries the AI search volume the index gave it, so the prompts table shows the figure straight away.
A row already in your portfolio reads Tracked, Suggested or Paused instead of offering the button, so you never add a prompt twice.
Markets and indexes
The Market picker offers the property's measured Markets plus its target Markets. A property with one Market has no picker.
The ChatGPT index covers the United States in English only, so it is offered only when the Search's Market is English (United States). Every other Market searches the Google AI Overview index. The Market and the index are part of the Search, so the same phrase in two Markets is two Searches, and two cached results.
Search history
The History button beside the tabs opens a sheet listing every Search the property has run: who ran it, when, and which version they saw. Open one and it joins the tabs, so you read the result your colleague read. Opening a past result never counts against the allowance.